Table 26 Context-specific deployment recommendations.
Clinical context | Recommended synthetic ratio (95% CI) | Prevalence weighting strategy | Expected performance range (95% CI) | Key considerations | Confidence level | Risk assessment score |
|---|---|---|---|---|---|---|
Population screening | 50–75% (45–80%) | Natural prevalence weighted | 84–89% accuracy (82–91%) | Balance sensitivity/specificity | High | 3/10 (low risk) |
Specialized clinics | 75% (70–80%) | Hybrid balanced-weighted | 91–96% accuracy (89–97%) | Maintain rare pathology detection | Very High | 2/10 (very low risk) |
Emergency settings | 50% (40–60%) | Cost-sensitive learning | 78–85% accuracy (75–87%) | Minimize false positives | Moderate | 5/10 (moderate risk) |
Research applications | 75% (70–80%) | Artificially balanced | 94–97% accuracy (93–98%) | Maximize overall performance | Very High | 1/10 (minimal risk) |
International deployment | 60–70% (55–75%) | Population-specific weighting | 82–88% accuracy (79–90%) | Account for prevalence variations | Moderate | 4/10 (low-moderate risk) |
Mobile screening | 50–60% (45–65%) | Robust focal loss | 79–86% accuracy (76–88%) | Handle extreme class imbalance | Moderate | 5/10 (moderate risk) |
